1. We used up to 35 years of capture-recapture data from nearly 3300 individual female ducks nesting on Engure Marsh, Latvia, and multistate modelling to test predictions about the influence of environmental, habitat and management factors on breeding dispersal probability within the marsh. 2. Analyses based on observed dispersal distances of common pochards and tufted ducks provided no evidence that breeding success in year t influenced dispersal distance between t and t + 1. 3. Breeding dispersal distances (year t to t + 1) of pochards and tufted ducks were associated with a delay in relative nest initiation dates in year t + 1. The delay was greater for pochards (c. 4 days) than for tufted ducks (c. 2 days) when females dispersed > 0.8 km. 4. Northern shovelers and tufted ducks moved from a large island to small islands at low water levels and from small islands to the large island at high water levels before the construction of elevated small islands (1960-82). Following this habitat management (1983-94), breeding fidelity was extremely high and not influenced by water level in the marsh for either species. 5. Because pochard nesting habitats in black-headed gull colonies were saturated during the entire study period, nesting females moved into and out of colonies with similar probabilities. Local survival probabilities and incubation body masses were higher for both yearlings (SY) and adults (ASY) nesting within gull colonies, suggesting that these females were of better quality than females nesting outside of the colonies. 6. Tufted ducks showed higher probabilities of moving from islands to emergent marshes when water levels were higher both before and after habitat management. However, rates of movement for a given water level were higher during the period before management than after. 7. Both pochards and tufted ducks exhibited asymmetric movement with respect to proximity to water, with higher movement probabilities to near-water nesting locations than away from these locations. 8. Multistate capture-recapture models provided analyses that were useful in investigating sources of variation in breeding dispersal probabilities.